Introduction

The diagnosis of Helicobacter pylori infection is based on either non-invasive methods—such as serological testing, the urea breath test, or stool antigen detection—or invasive approaches, including histology, culture, polymerase chain reaction (PCR), and next-generation sequencing (NGS) techniques, which analyze gastric or duodenal biopsy specimens1. The choice between non-invasive and invasive diagnostic tests depends on the clinical context. Histological examination has long been regarded as the gold standard for diagnosing H. pylori infection1. It not only assesses the extent of gastric inflammation but also enables the classification of lesions, particularly those that are ulcerative, metaplastic, dysplastic, or malignant, such as adenocarcinoma or mucosa-associated lymphoid tissue lymphoma. Histology is typically complemented by immunohistochemistry to detect H. pylori infection. This comprehensive diagnostic information is crucial for clinicians in the effective management of these infections. Biopsies submitted for histological examination require only fixation in formalin at room temperature, a straightforward and widely accessible procedure. However, although histology can provide a comprehensive diagnostic overview, it does not offer information on the antibiotic resistance profile of the infecting H. pylori strain. Although fluorescence in situ hybridization technology is used in a few countries to detect resistance2, it remains uncommon in routine clinical practice. As a result, clinicians often rely on empiric, or probabilistic, eradication strategies based on local resistance patterns, such as bismuth-containing quadruple therapy or concomitant therapy3,4.

When sending a gastric biopsy to a microbiology laboratory for H. pylori culture, specific and more stringent transport conditions are required to preserve bacterial viability. Ideally, the biopsy should be placed in a dedicated transport medium such as Portagerm Pylori, which allows for refrigerated transport at + 4 °C within 48 h5. Alternatively, immediate freezing in liquid nitrogen or on dry ice in a sterile tube can be employed, although this approach necessitates a well-coordinated logistical framework. The use of physiological saline is not recommended, as it does not reliably maintain the viability of H. pylori1. Culturing H. pylori demands specialized expertise due to the organism’s fragile nature and the requirement for selective culture media, some of which are commercially available. Despite these challenges, culture offers a key advantage: it enables antibiotic susceptibility testing, allowing clinicians to prescribe personalized eradication regimens based on the resistance profile of the strain 6,7. This targeted approach is generally better tolerated than empirical, or probabilistic, treatments, leading to improved adherence and a reduced risk of premature discontinuation. Ultimately, this enhances treatment success rates. Despite their value, H. pylori culture and PCR testing are not routinely performed in many countries due to numerous obstacles. The fragility of the bacteria requires strict preanalytical conditions for transporting biopsies. H. pylori culture is difficult and requires a certain level of expertise. Access to commercial PCR kits for detecting H. pylori is not possible in many countries. The lack of reimbursement for this highly sensitive test is a major obstacle in many countries. As a result, many gastroenterologists are satisfied with a histological diagnosis, which has long been considered the gold standard. The main idea behind this project is to use these paraffin-embedded biopsies to provide clinicians with microbiological data enabling them, if they wish, to prescribe targeted antibiotic therapy for H. pylori infections.

As an alternative to culture, gastric biopsies can be analyzed using molecular biology techniques. H. pylori PCR most commonly targets the 23S rRNA, particularly the V-loop region, which harbors mutations associated with macrolide resistance at positions 2142 and 21438. Several real-time PCR kits with excellent analytical performance are commercially available in many countries9,10. PCR is more sensitive than histology, which may underestimate 10–20% of infections11. However, its widespread use is hindered by limited reimbursement by national health systems, with only a few countries, such as France, currently covering its cost. NGS approaches using DNA extracted from gastric biopsies have also been developed and described by several research groups, including our own12,13,14. These methods enable comprehensive analysis of the bacterial resistome and, potentially, the virulome. Despite their advanced capabilities, the high cost and lack of reimbursement by healthcare systems remain major barriers to their broader implementation. Although PCR and NGS represent powerful tools in the modern diagnostic arsenal for H. pylori infection, both still require gastric biopsies to be sent to a microbiology laboratory. Unfortunately, in France—as in many other countries—clinicians often favor procedural simplicity and typically limit diagnostic evaluation to histological analysis alone.

The aim of this study was to evaluate the performance of the target enrichment followed by NGS amplification techniques previously described by our group14 for characterizing the H. pylori resistome and virulome in fresh gastric biopsies. Specifically, we investigated whether this approach could be applied to DNA extracted from paraffin-embedded biopsies routinely sent to histology laboratories. In a cohort of 30 biopsies that were histologically positive for H. pylori, we demonstrated that this technique yields results comparable to those obtained from parallel biopsies collected from the same patients and analyzed in microbiology laboratories using real-time PCR and culture with antibiotic susceptibility testing.

Results

Global evaluation of sequencing depth and coverage

To evaluate our method using clinical specimens, we tested 30 samples that were all positive for H. pylori by real-time PCR, with cycle threshold (Ct) values ranging from 25.8 to 37 (Table 1; Supplementary Table 1). Although most samples demonstrated sufficient sequencing quality for downstream analyses, several exhibited suboptimal base depth and gene coverage, rendering the detection of resistance markers, characterization of virulence-associated genes, and phylogenetic analyses incomplete or impossible. Specifically, samples 6, 8, 10, 14, 17, 21, 23, 27, 28, and 30 failed to achieve complete characterization due to inadequate sequencing quality in 2 to 5 targeted regions. Correlation analysis between average sequencing depth and real-time PCR Ct values (Fig. 1) revealed that all samples with Ct values below 30.0 met the criterion of achieving a mean depth of at least 10 × across all targeted regions (Supplementary Table 2). Notably, samples 11 and 18 displayed unexpectedly high sequencing quality despite their relatively late Ct values of 32.2 and 33.7, respectively. Overall, average sequencing depth across target regions decreased as H. pylori PCR Ct values increased (Fig. 1). A log–log linear regression model fitted to the data confirmed a significant negative association between Ct and average depth (p < 0.0001***), with an adjusted R2 of 0.66. Based on this model, we estimated a Ct cutoff of approximately 32.9, above which samples are unlikely to achieve an average sequencing depth of 10 × across the targeted regions.

Table 1 Description of included cases.
Full size table
Fig. 1
Fig. 1
Full size image

Correlation between average sequencing depth across targeted regions and H. pylori PCR Ct values.

Another approach to assess the quality of our libraries involved characterizing their DNA composition. By tracking information at various stages of the mapping-based pipeline, we could determine whether sequencing reads were filtered out before mapping—primarily due to poor quality or insufficient length—or whether they mapped to the human genome, to H. pylori (“on-target”), or remained unmapped (Fig. 2). Samples with a higher H. pylori colonization, indicated by Ct values between 25.8 and 32.9 (24 samples), exhibited a significantly higher percentage of reads mapped to the H. pylori reference genome (14 ± 4%) compared to samples with Ct values above 33, which showed only 3% ± 1% mapped reads (Fig. 2A). All samples with over 10,000 reads mapped to H. pylori demonstrated excellent sequencing quality (Supplementary Fig. 1). Although gene coverage correlated with the number of reads, it did not always reflect read quality. Several libraries were contaminated with human and unidentified DNA. For example, sample 30, which had one of the highest Ct values for H. pylori, contained 62% human DNA but only 1% of reads mapped to H. pylori. This likely indicates non-specific capture of unwanted DNA by probes that hybridize to conserved regions present in other bacteria—such as the 16S or 23S rRNA operons—or to the human genome. Additionally, some libraries, such as samples 17 and 23, showed a higher proportion of short fragments (< 50 bp) (Fig. 2A). This phenomenon may be caused by the presence of primer dimers, which typically arise when the targeted DNA content in the sample is low.

Fig. 2
Fig. 2
Full size image

Distribution of sequencing reads by category, illustrating sample quality.

A notable limitation of this approach is that a proportion of PCR positive samples, particularly those with higher Ct values, could not be fully characterized due to insufficient sequencing depth and coverage of all targeted regions. This likely reflects the limited amount of H. pylori-specific DNA relative to host background DNA, as well as potential library preparation artifacts such as probe homodimers.

Detection of clarithromycin resistance markers

Mutations in the 23S rRNA gene identified via target enrichment were compared to those detected by real-time PCR on fresh gastric biopsies. The concordance between the two methods was good in 25 of the 30 samples tested (83.3%), with all genotypes accurately identified (Table 2). The predominant mutation detected was A2143G, and the sole A2142C mutation (sample 8) was also correctly identified. However, base depth was insufficient in samples 6, 10, 21, and 23—each with Ct values ranging from 33.3 to 37 in DNA extracted from FFPE tissue (Table 1; Supplementary Table 1). A mixed population of wild-type and mutant H. pylori was detected in samples 18 and 27, with the mutant alleles representing 70% and 90% of the called bases, respectively. Notably, for sample 27, no mixed population was identified by AST, real-time PCR on fresh gastric biopsies, or WGS of the corresponding H. pylori isolate. This suggests a heterogeneous distribution of the clarithromycin-resistant subpopulation within the patient’s stomach, detectable only in the FFPE biopsy samples. When compared with AST results, the correlation with target enrichment data from FFPE samples was good. In sample 6, which had insufficient base depth by target enrichment, and sample 18, mixed populations were confirmed both in vitro, by real-time PCR on gastric DNA, and by WGS of the corresponding H. pylori strain.

Table 2 23S rRNA mutations linked to clarithromycin resistance detected among studied samples.
Full size table

Detection of levofloxacin resistance markers

Mutations in the quinolone resistance-determining region (QRDR) of the gyrA gene identified via target enrichment were compared with those detected by WGS of the corresponding H. pylori strains (Table 3). The correlation was good in 20 of the 30 samples (66.7%). Insufficient base depth was observed in nine samples (samples 6, 8, 14, 17, 21, 23, 27, 28, and 30), all but one (sample 28) testing positive by H. pylori real-time PCR with Ct values above 30 cycles. In sample 29, a mixed population was detected by AST and WGS, whereas target enrichment identified only a wild-type population. This suggests a heterogeneous distribution of the levofloxacin-resistant subpopulation in the patient’s stomach, detectable only by culture, despite the resistant bacteria constituting approximately 10% of the total population. The mutations associated with levofloxacin resistance identified were D91G (one sample with mixed infection), D91N (five samples, including one mixed infection), N87K (three samples), and N87I (three samples).

Table 3 GyrA mutations linked to levofloxacin resistance detected among studied samples.
Full size table

Detection of rifabutin and tetracycline resistance markers

Insufficient base depth for the rpoB gene was observed in the same nine samples as for gyrA (samples 6, 8, 14, 17, 21, 23, 27, 28, and 30). A H540Y mutation in rpoB was identified by target enrichment in sample 12; however, this mutation was not confirmed by either AST or WGS of the corresponding H. pylori isolate (Suppl Table 3). This discrepancy may again reflect a heterogeneous distribution of the resistant subpopulation within the patient’s stomach, detectable only in FFPE samples. No other mutations were detected in the remaining samples, consistent with the NGS results from the cultured H. pylori strains. No resistance markers for tetracycline were identified in any of the 30 samples (data not shown), as expected, since no tetracycline-resistant H. pylori isolates were included in this study. This outcome supports the reliability of sequencing for genes involved in tetracycline and rifampicin resistance. The only exceptions were the nine samples (6, 8, 14, 17, 21, 23, 27, 28, and 30) that exhibited poor sequencing quality across these targeted regions (Suppl Tables 1 and 4).

Virulence typing

The cagA gene was detected by target enrichment in 13 of the 14 cagA-positive samples (92.9%) identified by WGS of the corresponding strains (Table 4). Except for sample 30, the assembly quality of cagA sequences was sufficient to detect the EPIYA phosphorylation motifs in all other 12 samples. The determination of the number of EPIYA-C motifs was good, with the exception of sample 20, for which target enrichment identified two C motifs, whereas WGS of the H. pylori strain detected only one. Regarding vacA genotypes, the correlation between target enrichment and WGS was good in 22 of the 30 samples (73.3%). No genotype prediction was possible for samples 6, 21, 23, and 27 due to insufficient assembly length. For samples 10, 30, and 17, only the s1, i1, and i2 genotypes were identified, respectively. In sample 1, which was classified as s1i2m2 by WGS, the m genotype was not detected by target enrichment.

Table 4 Observed cagA and vacA genotypes from strain and paraffin-embedded sequencing.
Full size table

MLST of clinical samples

Geographic attribution of samples was determined from the concatenated sequences of the H. pylori typing scheme using STRUCTURE and a database of annotated sequences from PubMLST. The results obtained by WGS of H. pylori were compared with those from target enrichment (Table 5). The correlation between the two methods was good. Most biopsies (16 of 30, 53.3%) contained hpEurope H. pylori strains. Seven biopsies (23.3%) contained hpEurope/hpAfrica1 hybrids, with hpAfrica1 strains representing between 11 and 86% of the profile according to WGS results, or an estimated 7% to 75% by target enrichment. Six cases (20%) were attributed solely to hpAfrica1, and one remaining sample was identified as an hpEurope/hpEAsia hybrid. A minor discrepancy was observed for sample 10, which was classified as hpEurope by target enrichment but identified as an hpEurope/hpAfrica1 hybrid by WGS of the strain. Lastly, the geographic cluster for sample 23 could not be determined due to insufficient sequencing quality in the relevant regions of the typing scheme.

Table 5 Observed MLST profiles from strain and paraffin-embedded sequencing.
Full size table

Discussion

In this work, we demonstrate that target enrichment followed by NGS amplification of DNA extracted from FFPE biopsies enables comprehensive analysis of the H. pylori resistome and virulome. The concordance between NGS results and those obtained in vitro (PCR and antibiogram) on fresh biopsies was excellent.

Compared with our previous study on fresh biopsies14, coverage and read depth were improved. We also obtained analyzable results from DNA extracted from FFPE biopsies, positive by PCR, with Ct values close to 30. The fragmentation of DNA caused by formalin fixation in FFPE blocks appears to be an advantage for our technique, which relies on small RNA probes that can bind more effectively to degraded bacterial genomic DNA.

The sensitivity of target enrichment on FFPE samples is limited to those positive by PCR using our in-house assay with Ct values up to 30 cycles. Any laboratory wishing to implement this strategy will need to validate the detection limit based on their own routine PCR format.

In two cases included in this study, only the susceptible or resistant population were detected by target enrichment, unlike the results from in vitro testing. This discrepancy may be due to the fact that biopsies for histology and bacteriology not being taken from exactly the same anatomical site. In fact, more biopsies are taken for histological diagnosis than for bacteriology. The area of the mucosa covered by histology is therefore larger (a minimum of 5 biopsies is recommended), which increases the chances of identifying mixed infections in patients infected with strains with different resistance profiles and distributed heterogeneously in the gastric mucosa. Previous studies have shown that some patients may harbor mixed infections with respect to certain antibiotic resistance or virulence markers15,16.Our NGS approach on FFPE biopsies may therefore offer an advantage in detecting these mixed infections. The discrepancy we observed in a biopsy where 20% of a potentially rifamycin-resistant population was detected is particularly notable. It is likely that rifabutin treatment in this patient would not have been effective. This finding supports the recommendation to collect multiple biopsies from different anatomical sites to capture more accurately the bacterial diversity present. For levofloxacin, as our approach did not detect heteroresistance in sample 29, in which resistant bacteria represented approximately 10% of the population, it may suggest limited sensitivity for minority variants.

Other groups have described alternative NGS approaches applicable to DNA extracted from FFPE biopsies13. However, our study has the advantage of comparing target enrichment results with those obtained by real-time PCR, culture, AST, and WGS on all included biopsies.

Compared with commercially available kits such as American Molecular17 or GenoScreen Deeplex HelP (https://www.deeplex.com/deeplex-help/), our probe design offers greater adaptability, allowing the addition of resistance or virulence markers as new genomic markers are validated. For example, once markers for H. pylori resistance to amoxicillin or metronidazole are confirmed, corresponding probes can be incorporated. Our probe design has been updated from the version described by Gillet et al.14, notably by including the determination of the htrA genotype, recently identified by Prof. S. Backert’s team as an important virulence marker for H. pylori18,19. The resistome analysis on FFPE biopsies was highly effective, including accurate determination of the number of repeats of the C motif in CagA EPIYA phosphorylation sites20. The determination of vacA genotypes was also accurate; however, as with fresh biopsies, probe adjustments may be needed to improve coverage of the m2 genotype. The absence of patients with pre-neoplastic lesions or adenocarcinoma in our cohort likely explains why all patients infected with a cagA-positive strain had a wild-type htrA genotype. Additionally, the number of strains with C motif repeats was low.

It would now be valuable to test routinely a larger number of biopsies in collaboration with clinicians and pathologists to further demonstrate the clinical utility of these NGS results. Our center offers this technique on FFPE blocks specifically for cases of eradication failure, particularly in patients where successful eradication would provide significant clinical benefit. The turnaround time for this approach is approximately one week. Patients with ulcers, pre-neoplastic lesions, mucosa-associated lymphoid tissue lymphoma, or adenocarcinoma represent particularly relevant candidates for this new technology. With a view to routine use of the strategy proposed in our study, obvious precautions will need to be taken to avoid any cross-contamination between samples. In addition to these potential contamination risks, another potential challenge that could limit the adoption of NGS is its use in countries with limited economic resources. However, partnerships with expert laboratories such as ours, provide solutions for the routine application of this new approach.

A limitation of our approach is that approximately 30% of FFPE samples had insufficient coverage for some targets. The inherent fragmentation of FFPE-derived DNA may bias detection toward smaller genomic regions, potentially underrepresenting larger or more complex loci. Rare genotypes or low-frequency variants, such as certain htrA or vacA alleles, may therefore be under-detected. The limited performance of target enrichment in low-bacterial-load FFPE samples, also impacts resistance detection, virulence typing, and strain attribution. Finally, the absence of patients with pre-neoplastic lesions or adenocarcinoma also limits the generalizability of the findings to broader clinical populations.

Nevertheless, the use of target enrichment on DNA extracted from FFPE biopsies constitutes a promising approach with clear advantages. It offers clinicians a promising tool to implement targeted eradication strategies without requiring changes to their current clinical practices.

Methods

Biopsy

In total, 30 gastric biopsies submitted to the pathology department of Bordeaux University Hospital were analyzed. These samples were collected between 2021 and 2024 from 13 men and 17 women, with a median age of 38.6 ± 26.5 years (Table 1). The patients presented with various clinical conditions, including gastritis (n = 17), ulcers (n = 2), epigastric pain (n = 2), anemia (n = 1), emesis (n = 1), gastroesophageal reflux disease (n = 1), and other disorders (n = 6) (Table 1). H. pylori infection was confirmed in all cases by immunohistochemical staining using the Leica Bond III automated immunostainer protocol (Leica IHC-F, Leica Microsystems SA, Nanterre, France), with detection performed using the ULC3R clone antibody at a 1:50 dilution. In addition, all patients were confirmed positive for H. pylori through culture and real-time PCR analysis of fresh gastric biopsies sent in parallel to the National Reference Center for Campylobacters and Helicobacters (NRCCH). The real-time PCR used for H. pylori detection coupled with the main mutations associated with resistance to clarithromycin was previously described21. It corresponds to ready-to-use PCR microwell strips used on LightCycler® 480. These PCR microwell strips are available for purchase from Eurogentec (Liège, Belgium).

Bacterial culture

In total, 30 H. pylori isolates, corresponding to the same patients whose formalin-fixed paraffin-embedded (FFPE) biopsies were analyzed by target enrichment, were obtained from fresh gastric biopsies sent to the NRCCH. The biopsies were initially cultured on in-house prepared blood agar supplemented with antibiotics and incubated under microaerobic conditions in a controlled workstation (Baker Ruskinn, Concept Ruskinn, Bridget, UK) at 36 °C for up to 10 days. Antimicrobial susceptibility testing (AST) data were collected for all isolates. AST was performed on Mueller–Hinton agar supplemented with 10% sheep blood and freshly prepared globular extract. Minimum inhibitory concentrations (MICs) for clarithromycin and levofloxacin were determined using Etest® strips (bioMérieux), with breakpoint values established according to the guidelines of the French Microbiology Society’s Antibiogram Committee (CA-SFM/EUCAST) (CA-SFM / EUCAST: Société Française de Microbiologie Ed; 2024: p.1–177). MICs for clarithromycin, levofloxacin, rifampicin, and tetracycline were independently evaluated by two readers. Quality control was maintained using the H. pylori reference strain CCUG 17874.

Whole-genome sequencing (WGS) of H. pylori strains cultured from gastric biopsies

All 30 H. pylori isolates underwent WGS and AST. DNA extraction was performed using a combination of mechanical disruption (bead-beating), enzymatic digestion, and chemical lysis on Revvity’s Chemagic Prime instrument. Library preparation was conducted with Illumina’s DNA Prep kit utilizing Tecan’s DreamPrep NGS automation system, followed by sequencing on the Illumina NovaSeq 6000 platform. Bioinformatic analysis was carried out according to the pipeline described in our previous study, with the exception that genome assembly was performed using NCBI’s SKESA v2.5.122. An additional annotation step was included to extract genomic sequences of the cagA and vacA genes using Prokka v1.14.523.

Probe design for target enrichment

An RNA probe library comprising 13,245 probes of 120 nucleotides each, totaling 108 kbp, was designed by Agilent (Santa Clara, CA, USA). This probe library enables the detection of genomic mutations associated with antimicrobial resistance to clarithromycin (23S rRNA), levofloxacin (gyrA), rifampicin (rpoB), and tetracycline (16S rRNA) (Table 6)14. In addition, the library targets all genomic regions commonly used in the H. pylori multilocus sequence typing (MLST) scheme, as described by Falush et al.24,25. Probe design can be ordered from Agilent by referring to our laboratory.

Table 6 List of mutations associated with resistance to primary antibiotics.
Full size table

Automated capture-based library preparation

Most steps of the SureSelectXT workflow were automated using Agilent’s MagnisDx library preparation system, covering the process from enzymatic fragmentation to hybridized library capture. DNA extracted from FFPE biopsy samples using the QIAamp DNA FFPE Tissue Kit, was normalized to 100 ng in a 14 µL input volume. The size of the DNA extracted from the FFPE samples ranged from 200 to 1000 bp. Following Agilent’s guidelines, enzymatic fragmentation was performed for 20 min for high-quality DNA. The protocol included 18 pre-capture PCR cycles and 24 post-capture PCR cycles.

NGS short-read sequencing

The final library fragment size distribution and molarity were assessed using Agilent’s TapeStation 4150 system with the High Sensitivity D1000 assay. A 2 µL aliquot of the library was diluted tenfold, mixed with 2 µL of High Sensitivity D1000 Sample Buffer, and normalized to an equimolar concentration of 0.125 nM. A total of 20 µL of this final mixture was then prepared for multiplex sequencing, which was carried out on the Illumina iSeq 100 sequencer (San Diego, CA, USA) with a sequencing read length of 2 × 150 base pairs and paired-end sequencing, at the NRCCH.

Bioinformatics workflow

Raw sequencing reads were initially mapped to the human genome using Bowtie2 v2.4.5 with the Homo sapiens GRCh38 reference assembly. Unmapped reads were subsequently trimmed and cleaned using fastp v0.23.4. To reduce false positives, a deduplication step exploiting duplex metabarcoding of inserts was applied to generate consensus sequences. These processed reads were then aligned to the H. pylori reference genome J99 (assembly ASM98269v1) using Bowtie2. Consensus sequences for all targeted regions were extracted with FreeBayes v1.3.6 and BCFtools v1.19, with a minimum base depth required of 5X. A custom bioinformatics pipeline was then employed on these consensus sequences to detect genomic resistance markers and characterize the bacterial resistome.

Concurrently, a de novo assembly was performed using NCBI’s SKESA v2.5.1 to reconstruct complete sequences of the cagA and vacA genes. Sequences corresponding to the MLST scheme were concatenated and analyzed with STRUCTURE v2.3.4 to assign each sample to a defined population cluster. These sequences were then compared against the continuously updated allele sequence database available on PubMLST. Phylogenetic trees were generated using RAxML-NG v1.2.2 with the GTR + Gamma substitution model. Finally, a set of in-house scripts was developed to integrate all analysis outputs, generate PDF and HTML reports, and produce circular genome visualizations using the Circos graphic library v0.69–9. FastQC v.0.12.1 and QUAST v.5.2.0 are used to assess mapping and assembly quality respectively.

Statement

All diagnostic methods were performed routinely in accordance with relevant guidelines and regulations. All patients were investigated in a hospital setting, according to good clinical practices. In this routine process, consent for the endoscopic procedure and biopsy collection is always provided and kept in the patient’s medical record.